JP2002123544A - Retrieval preprocessing apparatus, document retrieving apparatus, and retrieval preprocessing method and document retrieving method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a retrieval preprocessing apparatus and a document retrieving apparatus, and a retrieval preprocessing method and a document retrieving method which make it possible to easily create a proper retrieval sentence and enable retrieval wherein retrieval intention is properly reflected by generating a natural retrieval sentence by using a candidate word and performing retrieval. SOLUTION: Candidate words as candidates for retrieval conditions are presented and a candidate word that a retriever selects out of the candidate words is employed as a 1st word for a retrieval sentence; and candidate words following the word employed for the retrieval sentence are presented and a candidate word that the retriever selects out of the candidate words is employed for the retrieval sentence and added. The presentation of a candidate word following a word employed for the retrieval sentence and the addition of the candidate word selected by the retriever to the retrieval sentence are repeated as many times as needed to generate the retrieval sentence, an object document is retrieved on condition that the retrieval sentence is included, and the retrieval result is presented.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention A search pre-processing apparatus, a document search apparatus, a search pre-processing method, and a document search for performing a search using natural words so that a search that reflects the intention of a searcher easily and accurately can be performed. About the method.

[0002]

2. Description of the Related Art Searching in a natural language requires complicated processing such as synonym processing and convergence of different notations in addition to sentence analysis in order to correctly judge the search intention from an indefinite document freely described by the searcher. It is.

[0003] In the current natural language search, words considered to be effective in the search are extracted by morphological analysis of the search sentence.
It is common to perform a match search as a character string.

[0004]

However, in this method,
Since the searcher does not know the content of the document to be searched and the words used therein, there is a problem that it is difficult to create an appropriate search sentence expressing the search intention.

To compensate for this, synonyms and related words may be presented together with the search results. However, these words are extracted based on limited words used in the search, and are not necessarily searched. It is not in the context of the sentence.

[0006] Further, since the causal relationship between the search condition and the result is not clear, if the desired result cannot be obtained,
Another problem is that it is not known whether the search sentence should be corrected or the target document is not included in the search target.

Therefore, the present invention makes it possible to easily create an appropriate search sentence by creating a natural search sentence using candidate words and performing a search, and the search intention is appropriately reflected. It is an object of the present invention to provide a search preprocessing device, a document search device, a search preprocessing method, and a document search method that enable a search.

[0008]

To achieve the above object, the present invention employs the following means.

The document search apparatus and the document search method of the present invention perform a search using a natural search sentence combining candidate words, and enable a search in which a search intention is appropriately reflected.

Further, the pre-searching apparatus and the pre-searching method of the present invention extract a candidate word from a search target document, thereby making it possible to generate an appropriate search sentence using natural words contained in the search target document itself. I can make it.

Therefore, it is possible to reduce the number of times of trial and error without matching the searcher's intention with the search result.

[0012] A search preprocessing apparatus according to the present invention divides a search target document into words in a predetermined unit and creates candidate words serving as search condition candidates, and uses the candidate words as appearances of the candidate words. A candidate word storage means for storing in association with the form.

In this manner, the search target document is defined as a word in a predetermined unit,
For example, by dividing a word, a phrase, and a part of speech to urge a candidate word, and storing which candidate word appears next to which candidate word, that is, by storing the appearance form in association with the candidate word,
When creating a search sentence, the appearance form can be reproduced, and a natural and appropriate search sentence can be created.

In this case, among the candidate words created by the candidate word creating means, an integrating means for integrating the number of occurrences of the same candidate word and a number storage means for storing the number of occurrences for each candidate word are provided. May be configured.

[0015] Candidate words having a high appearance frequency tend to be frequently used when creating a search sentence. Therefore, when creating a search sentence, presenting the candidate words in descending order of the number of appearances is convenient. Is improved.

[0016] The apparatus may further comprise candidate word weighting means for performing weighting according to the importance of the candidate words, and weight storage means for storing weighting information for each of the candidate words.

In this case, it is preferable that the importance of the candidate word is at least one of the number of appearances and the appearance location of the candidate word.

As described above, by weighting the candidate words, when the candidate words are presented, the candidate words can be presented in descending order of importance, and the convenience is improved.

Further, the above-mentioned search preprocessing apparatus may be configured to include a keyword extracting means for extracting a keyword from a candidate word and a keyword storing means for storing the keyword.

In this case, it is preferable to provide a combination creating means for creating a combination of the keywords present in the search target document, and a combination storage means for storing the combination of the keywords in association with the document in which the combination exists. It is.

[0021] The document search apparatus of the present invention searches for a plurality of documents as search targets and searches for documents that match the search conditions, and presents candidate word presenting means for presenting candidate words that are candidates for search conditions. A search sentence creating means for creating a search sentence by combining candidate words presented by the candidate word presenting means, and a search result presenting means for presenting a result of searching for the search target with the inclusion of the search sentence as a search condition. Is provided.

As a result, an appropriate search sentence combining candidate words can be created, and a search in which a search intention is appropriately reflected can be performed.

In the above-described document search apparatus, it is preferable that the search sentence creating means employs, as the search sentence, a candidate word selected by a searcher among the candidate words presented by the candidate word presenting means. For example, the searcher selects a candidate word displayed on the display by the candidate word presenting unit with a mouse or the like, and the search sentence creating unit adds (adopts) the candidate word to the search sentence. By repeating this selection and addition, a search sentence in which several candidate words are linked is created and searched. This makes it possible to easily create a search sentence according to the searcher's intention.

Further, the candidate word presenting means is configured to divide the search target document into words of a predetermined unit and generate the search text based on the appearance of the candidate word in the search target document. The candidate word following the adopted candidate word is presented to the creating unit, and the search sentence creating unit adopts the candidate word selected by the searcher among the following candidate words and adds the candidate word to the search sentence. It is preferable to adopt a configuration.

Thus, the candidate words included in the search target document can be presented based on the appearance form, and the search sentence can be easily created.

Further, the candidate word presenting means, in accordance with the order of the plurality of candidate words employed prior to the search sentence creating means, causes the next succeeding candidate word in the document to be searched for the candidate word. It is good also as a structure determined and presented based on the appearance form.

With this arrangement, it is possible to present candidate words in a context based on the order of a plurality of candidate words.

[0028] The candidate word presenting means may present the candidate words in an order according to the number of appearances of the candidate words in the search document.

Since a candidate word having a high frequency of appearance tends to be frequently used when creating a search sentence, it is convenient to present the candidate words in the order of appearance frequency when creating a search sentence. Is improved.

[0030] The candidate word presenting means may present the candidate words in an order according to the importance of the candidate words.

As described above, by weighting the candidate words, the candidate words can be presented in the order of importance when presenting, and the convenience is improved.

[0032] Of the plurality of candidate words adopted by the search sentence creating means, a noun or a noun phrase may be combined into a search sentence.

[0033] Among the plurality of candidate words employed by the search sentence creating means, a noun or a noun phrase is combined into a search sentence, and the search result presentation means divides the search target document into words of a predetermined unit and creates the search sentence. The combination of keywords that match the search sentence is searched for by referring to the combination of keywords that are nouns or noun phrases among the candidate words that have been selected, and a document that includes the combination of keywords that matches the search sentence is presented as a search result. It is good also as composition. At this time, when there are a plurality of combinations that are search sentences, a sentence including all of these combinations may be extracted as a search result, or a sentence including any one of the combinations may be extracted as a search result. May be.

The search preprocessing method of the present invention divides a document to be searched into words of a predetermined unit, creates candidate words as search condition candidates, and associates the candidate words with the appearance forms of the candidate words. Remember.

As described above, by storing candidate words in association with appearance forms, when creating a search sentence, the appearance form can be reproduced, and a natural and appropriate search sentence can be created.

In this search preprocessing method, for example, a search target document is divided into words of a predetermined unit, candidate words serving as search condition candidates are created, and the candidate words are stored in association with the appearance forms of the candidate words. May be provided as a recording medium on which a program for causing a computer to execute the processing to be performed is recorded.

Further, the document search method of the present invention presents a candidate word as a search condition candidate, employs a candidate word selected by a searcher among the candidate words as a first word in a search sentence, Presenting candidate words following the word adopted in the search sentence,
A candidate word selected by the searcher from among the candidate words is adopted and added to the search sentence, a candidate word following the word adopted in the search sentence is presented, and a search sentence of the candidate word selected by the searcher is added. A search sentence is created by repeating the addition and addition to the required number of times, and a search target document is searched using a search condition that the search sentence is included,
Present search results.

Thus, an appropriate search sentence combining candidate words can be created, and a search in which the search intention is appropriately reflected can be performed.

In this case, the search target document is divided into words in a predetermined unit to create candidate words, and the candidate words are stored together with the appearance form in the search target document, and the search word is stored based on the appearance form. It is preferable to determine candidate words that follow the word employed in the sentence.

This document search apparatus presents, for example, candidate words which are candidates for search conditions, and employs, as a first word, a candidate word selected by a searcher from among the candidate words in a search sentence. Presenting a candidate word following the word adopted in the search word, adopting and adding to the search sentence a candidate word selected by the searcher from the candidate words, and presenting a candidate word following the word adopted in the search sentence And the addition of the candidate word selected by the searcher to the search sentence is repeated as many times as necessary to create a search sentence. The search target document is searched using the search sentence being included as a search condition, and the search result is presented. May be provided as a recording medium on which a program for causing a computer to execute the processing to be performed is recorded.

In the present invention, the above components can be combined as much as possible.

[0042]

DESCRIPTION OF THE PREFERRED EMBODIMENTS <First Embodiment> An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a conceptual diagram of a search preprocessing device and a document search device according to an embodiment of the present invention. According to the present invention, a candidate word that is a candidate for a search condition is presented, and a search is performed using a natural search sentence combining the candidate words.

The search preprocessing apparatus 1 divides a document to be searched into words of a predetermined unit and creates candidate words serving as search condition candidates, and a candidate word created by the candidate word creation means 1a. Among the words, there are provided an accumulating means 1b for accumulating the number of occurrences of the same candidate word, and a keyword extracting means 1c for extracting a keyword from the candidate word.

A candidate word storage means (candidate word table) 1f for storing the candidate words in association with the appearance forms of the candidate words, and a number storage means (number of times) for storing the number of appearances of the candidate words for each candidate word Table) 1g, keyword storage means (keyword table) 1 for storing the keyword
h.

Although these storage means 1f to 1h may be provided independently of each other, they are stored as tables of the database 10 in this embodiment. Note that these storage means (database 10) may be provided on the document search device side. The search target document is stored in the database 10 or a separate database (including a case where the document is stored in each server on the Internet).

On the other hand, the document search device 2 combines a candidate word presenting means 2a for presenting a candidate word as a search condition candidate and a candidate word presented to the candidate word presenting means 2a to create a search sentence. It comprises a sentence creating means 2b and a search result presenting means 2c for presenting a result of searching for a search target with the inclusion of the search sentence as a search condition.

FIG. 2 is a flowchart of a search procedure performed by the document search apparatus. When the search is started, a keyword is searched for and searched to determine the first candidate word (S1). In this example, xDSL is to be checked. As shown in FIG. 3, "x" is input in a search box for "search for", and a "search for" button is selected, that is, clicked with a pointing device.

The candidate word presenting means 2a includes a database 10
And displays a list of keywords starting with this character (S2). See FIG.

Then, the searcher selects a keyword displayed as a list (S3). In this example, in order to search for xDSL, a natural language in the “xDSL” column is selected.

FIG. 5 shows an example of a search screen in natural language. The search sentence creating means 2b adopts the candidate word “xDSL” selected in the previous step as the first word of the search sentence, and the candidate word presentation means 2a outputs the candidate word table 1f
Is displayed, and the next candidate word that seems to follow this candidate word is displayed (S4).

In this example, the candidate words are displayed five by five,
When there are five or more candidates, the next five candidates and the previous five candidates are displayed using the ▽ and △ buttons.

When the searcher selects the next candidate word from these, the search sentence creating means 2b adds the selected candidate word to the search sentence, and the candidate word presentation means 2a returns to the candidate word table again. With reference to 1f, a candidate word determined to be continued next is displayed based on the appearance form of the candidate word.

The searcher repeats the selection of candidate words until a desired search sentence is created (S5). FIG. 6 shows a case where “no” is selected as the second candidate word.

In this example, "information" is used as the third candidate word.
Is selected, and the search button is selected in a state where the search sentence is “xDSL information”.

The search result presenting means 2c searches the document containing the search sentence from the search target document, and displays the search result in FIG.
(S6).

As described above, in the present embodiment, since the search is performed using the natural search text, the search intended by the searcher can be easily performed. In addition, since a search sentence is created while referring to candidate words obtained by processing a search target in advance, an appropriate search sentence can be easily created.

<Pre-Search Processing> Next, a procedure for creating a candidate word, which is a pre-search processing, will be specifically described. The candidate word creating means 1a first performs [separation processing] on the text shown in FIG. This segmenting process performs substantially the same operation as a conventional person, and divides a target document into word units to generate candidate words. In this example, when a title is attached to a document, a separate writing process is also performed on the title (see FIG. 9).

Each candidate word is stored in association with the appearance form. That is, when creating a search sentence, each candidate word is stored in the candidate word table together with the immediately preceding candidate word so that the next candidate word can be presented. FIG. 10 is an example in which the candidate words of FIG. 9 are stored in the candidate word table 1f. At this time, parentheses, symbols, spaces, and punctuation that are not appropriate for the candidate word are ignored.

When "(") is a noun like "dedicated workstation (high-performance personal computer)", the noun immediately after "(", the high-performance personal computer and the workstation are in the same line. It is better to register "no" as a candidate word in front of a high-performance personal computer, and the next word after ")" is "workstation before"(" Both words immediately before are pre-candidate words.

The integrating means 1b calculates one sentence, for example, from the beginning of the sentence to a period, for the candidate word in FIG.
The number of occurrences of the same candidate word in a group of sentences, such as a sentence enclosed in parentheses and a headline changed, is integrated, and the number of appearances for each candidate word is stored in the count table 1g (FIG. 1).
1).

Those with a large number of occurrences and a high frequency of appearance are:
Since the probability of being used for a search sentence is considered to be high, when the candidate word presenting means 2a presents the candidate word, the candidate word table 1
The next candidate words extracted from f are presented in order of appearance frequency with reference to the frequency table 1g.

The keyword extracting means 1c extracts a noun or a nominative from the candidate words in FIG. 9 and registers them in the keyword table 1h of the database 10 as keywords (see FIG. 12).

When the candidate word presenting means 2a searches for a keyword in step S1, this keyword table 1
The keyword is presented with reference to h.

<Second Embodiment> In this embodiment, instead of searching for the first candidate word in comparison with the above-described embodiment, a technical field is selected and the first candidate word is selected from among related candidate words. The difference is that the candidate words are presented, and the other configurations are substantially the same. In this example, the same elements as those in the above-described embodiment are denoted by the same reference numerals, and the description thereof will not be repeated.

FIG. 13 is a schematic configuration diagram of the present embodiment.
The keyword extracting means 1c 'of the present embodiment classifies keywords into related technical fields as described later and stores them in the keyword table 1h' in association with strongly related words.

The weighting means 1d performs weighting according to the importance of the keyword, stores the weight information in a weight table 1j of the database 10, and when the candidate word presenting means 2a presents the keyword, The more important keywords are, the more they can be presented first.

FIG. 14 is a flowchart of a search procedure according to this embodiment. When the search is started, first, in order to narrow down the field to be searched, the field to be searched is displayed on the display means such as a monitor by the candidate word presenting means 2a and presented to the searcher (S11). FIG. 15 shows a display example of the search field at this time.

The searcher selects a field to be searched from these fields (S12).

The candidate word presenting means 2 a
And displays a list of keywords belonging to the selected field (S13). FIG. 16 is a display example when "next-generation heart @ ware architecture" is selected.

Then, the searcher selects a keyword displayed as a list (S14). For example, when it is desired to search for xDSL, a natural language in the column of “xDSL” is selected.

FIG. 17 shows an example of a search screen in a natural language, in which five keywords having high relevance to "xDSL" are displayed in the candidate word column (S15). If there are five or more candidates, the next five candidates and the previous five candidates are displayed using the ▽ and △ buttons.

The searcher selects a candidate word from these (S16). The search sentence creating means 2b adopts the selected candidate word as the first word of the search sentence, and the candidate word presenting means 2a refers to the candidate word table 1f to determine the next candidate word following the candidate word. The candidate words are displayed (S17).

The searcher repeats the selection of candidate words until a desired search sentence is created (S17, S18). FIG. 6 shows a case where “no” is selected as the second candidate word. The search sentence creating unit 2b adds the selected candidate word to the search sentence, and the candidate word presentation unit 2a displays the next candidate word.

Similarly, the searcher selects “information” as the third candidate word, and selects the search button in a state where the search sentence is “xDSL information”.

The search result presenting means 2c searches the document containing the search sentence from the search target document and displays a list as shown in FIG. 7 (S19).

As described above, in the present embodiment, since the search is performed using the natural search text, the search intended by the searcher can be easily performed. In addition, since a search sentence is created while referring to candidate words obtained by processing a search target in advance, an appropriate search sentence can be easily created.

<Pre-Search Processing> Next, step S13
In steps S15 to S15, a procedure for extracting candidate words having high relevance for each keyword in order to present candidate words having high relevance to the selected keyword will be described.

The extraction of the candidate words is carried out in advance by a human reading a document to be searched or a general document, selecting words that are considered to be closely related to each keyword, and recording a related word for each keyword. May be used, but in this example, the inter-keyword distance based on the affinity measure of the keyword extracted from the search target document is obtained, and highly relevant candidate words are automatically extracted according to the distance. .

First, the document to be searched is divided into sentences of a predetermined unit (a set of sentences such as a sentence from the beginning of the sentence to a punctuation mark, a sentence enclosed in parentheses, a single paragraph, a headline changed, etc.). Extract keywords in the document.

Next, the number of unit documents containing a certain keyword x is integrated, and this is added to the appearance frequency N (x) of the keyword x.
And Similarly, the number of unit documents including another keyword y is integrated, and this is set as the appearance frequency N (y) of the keyword y.

Further, the number of unit documents including both the keyword x and the keyword y is defined as the co-occurrence frequency of the keyword (x, y), and the keyword distance L (x, y) is obtained according to the following equation.

L (x, y) = {N (x) + N (y) -N
(X, y) / N (x, y) It is determined that the smaller the distance L (x, y), the higher the relevance between the keyword x and the keyword y.

The distance L (x, y) is obtained for all combinations of keywords, and the keywords are stored in the keyword table 1 by associating each keyword with a keyword having a distance equal to or less than a predetermined value or up to a predetermined keyword in ascending order of distance.
h '.

Thus, the candidate word presenting means 2a can present candidate words highly relevant to the keyword by referring to the keyword table 1h '.

Next, in steps S11 to S13, in order to display keywords belonging to the selected technical field,
A procedure for classifying a keyword into a predetermined technical field will be described.

For the keyword extraction, a human may classify each keyword appropriately in a predetermined field in advance. In this example, however, a knowledge dictionary of the classification is created in advance using a known document, and a search target document is created. Were automatically classified based on this knowledge dictionary.

FIG. 18 is an explanatory diagram of a procedure for automatically classifying keywords. The keyword pairs that appear in many texts in a category are "Speaking A" in the category.
It can be said that the strength of this semantic connection directly expresses the concept of a set organized as a classification. In each category, by quantifying the strength of analogy of keyword pairs that appear in the category and accumulating it as knowledge, when a new text is given,
By comparing the keyword pairs appearing there with the knowledge, it is possible to determine which classification concept the text is closest to from the difference in analogy strength.

1. Construction of classification knowledge dictionary (1) Keyword extraction and co-occurrence creation Extract keywords and their combinations (co-occurrence) for all texts in the classification. Co-occurrence within a sentence for texts with long sentences such as papers and newspaper articles, and co-occurrence for the entire text for sentences that are short and not always grammatically correct, such as customer voices For example, the co-occurrence method is changed depending on the characteristics of the text to be used as a teacher to extract more accurate information.

(2) Distance Calculation The distance is calculated from the frequency of each of the two keywords forming the co-occurrence and the frequency of the co-occurrence using the equation (a). The larger the co-occurrence with the other in the frequency of each occurrence in the classification, the higher the probability that the other keyword will be in the same text if the keyword is present, so the distance between the two keywords is defined to be closer. I do.

(Equation 1)

(3) Calculation of Dependency Based on the above distance, the strength of analogy between keywords in the classification (degree of dependency) is defined by equation (b). The minimum value of the dependence (the dependence on the maximum distance Lmax in the classification) is 0.0
A constant is set for each classification so that it becomes 5, and the difference in numerical values between classifications due to the number of texts and the like is eliminated.

(Equation 2)

2. Automatic classification (1) Keyword extraction and co-occurrence creation Extract keywords and their combinations (co-occurrence) for new classification target text.

(2) Selection of Candidate Classification A classification that matches at least one co-occurrence extracted from the classification target text by collating with the knowledge dictionary for classification is selected as a corresponding classification candidate.

(3) Extraction of Dependency The combination of the keyword of the classification dictionary and the keyword of the text to be classified is collated, and the maximum dependence of the combination with the same element word in the candidate classification is extracted. In addition, for a keyword whose corresponding combination is not included in the candidate classification, the degree of dependence is set to 0.5.

(4) Weighting calculation A weighting parameter is obtained from the number of occurrence classifications (feature values) of the keyword and weighted. First, a weight parameter is obtained from the total number of classifications and the feature value by using the equation (c). Since the smaller the feature value is, the more the keyword is specific to the classification, the larger the weight parameter is.

(Equation 3)

Next, a fuzzy measure is calculated. At this time, especially for those having a small feature value, an addition point calculation is performed using the equation shown in the figure below. As a result, it is possible to give a difference in weighting between a keyword specific to a classification and a general keyword having many occurrence classifications. Which formula is used determines an appropriate threshold according to the total number of classifications.

(Equation 4)

(5) Classification effective value calculation The weighted calculation values of the fuzzy measure for all the keywords extracted from the text to be classified are added, and the average thereof is obtained as the classification effective value.

(Equation 5)

(6) Classification The text to be classified is classified as having the highest classification effective value.
If one classification target text may be classified into a plurality of texts, for example, a threshold of the classification effective value may be provided and classified into classifications having the threshold or more, or a predetermined number of classifications may be performed in descending order of the classification effective value. I decided to.

Next, the procedure of weighting by the weighting means 1d will be described. It focuses on the notational characteristics of sentences and extracts key keywords. Specifically, (1) important words indicating the contents of a sentence are often included in the subject and the object. (2) The title simply shows the contents of the sentence.

(3) The more important a word is, the more frequently it is described in a sentence. It is determined whether or not the keyword is an important keyword based on the characteristic described above. FIG. 19 shows a specific calculation method.

First, keywords are extracted from the target document in the same manner as described above. It should be noted that what the particle following the extracted noun is, and whether there is """" before and after the emphasis keyword are also stored.

Next, the following basic points are given to each of the extracted keywords.

That is, the basic point is determined by the word following the keyword. Specifically, if the word “ga” or “ha” indicates the subject phrase after the keyword, the basic point of the keyword is 0.8. And

Also, 0.7 is given as a base point for "wo" and "to" representing object phrases, and 0.7 for "de", "ya" and "mo". I do.

Furthermore, the keywords included in the title, which are described in the text, are set to 0.9, and those not described in the text are set to 0.6. Point.

The basic points of keywords that do not correspond to the above are all set to 0.5.

Also, since the more important keywords appear in the document many times, points that appear more than once are added according to the following formula (see 1-3 in FIG. 19).

V = (V0 + V1)-(V0 × V1) (A) where V0 is the basic point of the keyword that appears for the first time,
V1 is the basic point of the keyword that appeared for the second time,
The required V is the importance of the keyword. If the same keyword appears three or more times, the obtained V is calculated as V0 in equation (a), and the basic point of the newly appeared keyword is calculated as V1, and thereafter, the required number of times is repeated. .

A keyword with a suffix is deducted (see FIG. 1).
9-4). That is, the basic point is multiplied by a certain numerical value (a positive number equal to or less than 1 and referred to as a “deduction coefficient”).

As the suffix, for example, “name, quantity,
Wind, plan, diagram, table, chemical, system, sphere, material, person, species, number, product,
Theory, side, genus, value, target, degree, cost, part, law, use, faction, ratio,
Rate, flow, column, example, theory, drawing, group, type, column, point, gender, day,
Time, kind ".

However, some of the keywords with the suffixes as described above can be important keywords, for example, "female".

Therefore, in the case of two characters including the suffix, the number of dark spots is reduced.

Specifically, the number of characters is “2”; V ′ = V0 ×
0.7 Others: V ′ = V0 × 0.5 (where Vo is a basic point of each word).

[0113] Hiragana-mixed keywords are extinguished (see 5 in FIG. 19). The dark spot coefficient at this time is “0.
5 ".

The one-letter kanji disappears (see 6 in FIG. 19).

That is, although a word such as "what" is not suitable as a keyword,
There is a word that is added (the basic point is higher).

The words such as “what” and “next” have a high frequency of appearance, and the connecting particles are often “ga” and “wo”. In this way, in order to keep the weight of a phrase that cannot be an important keyword (useless) at a low level, a single character vague character is set as a vanishing point, and the vanishing point coefficient at that time is set to “0.
7 ".

The weight information of each keyword calculated as described above is as shown in FIG.

Thus, when displaying the keywords in step S13, the candidate word presenting means 2a refers to the weight table 1j and presents the important keywords first.

<Third Embodiment> This embodiment is different from the above-described embodiment in that when presenting a candidate word, it reflects the context of a search sentence created by the word selected earlier. The difference is that the next candidate word is extracted in consideration of the continuation of a plurality of candidate words, and other configurations are substantially the same. In this example, the same elements as those in the above-described embodiment are denoted by the same reference numerals, and the description thereof will not be repeated.

In this example, in the pre-search processing apparatus 1, the candidate words created by the candidate word creating means 1a are stored in the candidate word table 1.
When the candidate words are stored in f, a predetermined number of arrangement orders of the candidate words are stored. In this example, as shown in FIG. 21, one candidate word is immediately preceding the candidate word n−1, two previous candidate words n−2,
The third previous candidate word n-3 is stored.

Thus, when the candidate word is presented by the candidate word presentation means 2a of the document search device 2, the preceding three candidate words are referred to, and the candidate word in accordance with the context of the document to be searched can be created.

For example, a candidate word is created from the sentence in FIG.
As shown in FIG. 22, when “ID for telephone” is selected in advance, if the reference number is 1 as in the above-described embodiment, “of”, “at”, and “to” are presented. If the number of references is 3, as in this example, the number is reduced to "no" and "de". Although one candidate word of the search target document is shown for convenience, the candidate word is actually created from a plurality of documents, and the candidate word is presented by referring to this.

In this example, when the number of previously selected candidate words has not reached the number of references, the number of displayable reference words is reduced to the number of references that can be displayed. More specifically, when selecting the second candidate word following the first keyword, the number of references is set to 1, and "ni", "digital", and "but" are presented.

Similarly, when selecting the third keyword, the number of references is set to 2 and an “ID card” is presented.

<Fourth Embodiment> This embodiment is different from the second embodiment in that a search is performed using a combination of keywords, and other configurations are substantially the same. In this example, the same elements as those in the above-described embodiment are denoted by the same reference numerals, and the description thereof will not be repeated.

FIG. 23 is a schematic configuration diagram of the present embodiment.
The combination creating means 1e of the search preprocessing device 1 creates a combination of keywords existing in the document to be searched and stores the combination in the combination table (combination storage means) 1k of the database 10 in association with the document in which the combination exists. I have.

First, the combination creating means 1e outputs a sentence of a predetermined unit (a sentence from the beginning of a sentence to a punctuation mark, a sentence enclosed in parentheses, a single paragraph, a paragraph break) from among the keywords extracted by the keyword extracting means 1c '. Combine multiple items included in a group of sentences (headings, etc.). FIG.
FIG. 9 is an explanatory diagram in the case where a combination of two keywords is created by using a unit from the beginning of a sentence to a period as a unit.

The search sentence creating means 2b 'creates a natural search sentence by the searcher's selection as described above, and when the search button is selected, creates a combination of nouns from the search sentence.
This combination is passed to the search result presentation means 2c '. FIG. 25 is an explanatory diagram in the case where "phone having a plurality of extension numbers" is employed as a search sentence.

The search result presenting means 2c 'searches the combination table 1k for the combination shown in FIG. 25, and presents a target document in which this combination exists as a search result.

Since there is some relationship between the keywords appearing in the same sentence in expressing the meaning of the sentence, it is better to search by combination rather than searching by keyword alone for the purpose of the sentence “search sentence”. Is reflected in the search. That is, according to the present embodiment, a highly relevant document can be efficiently searched.

<Fifth Embodiment> FIG. 26 shows an example of a configuration for causing a computer 20 to execute a search preprocessing method and a document search method according to the present invention.

The computer 20 includes input means 21 such as a keyboard 21a and a pointing device 21b, storage means 22 for storing programs, search data, and the like, an arithmetic processing unit 23 for performing processing in accordance with the programs, and a display for displaying search results and the like. Means (output means) 24.

The recording medium 2 is stored in the computer 20.
5, the arithmetic processing unit 23 and the storage unit 22 constitute the candidate word creation unit 1a, the accumulation unit 1b, the keyword extraction unit 1c, and the database 10, Pre-search processing, that is, a process of dividing a search target document into words in a predetermined unit, creating candidate words as search condition candidates, and storing the candidate words in association with the appearance forms of the candidate words is executed. .

Further, by causing the computer 20 to read a program for executing the document search method from the recording medium 25, the arithmetic processing unit 23 and the storage unit 2 are read.
2. The display means 24 constitutes the candidate word presenting means 2a, the search sentence creating means 2b, and the search result presenting means 2c, and presents a document search process, that is, presents a candidate word as a candidate for a search condition. The candidate word selected by the searcher from the words is adopted as the first word in the search sentence, and the candidate word following the word adopted in the search sentence is presented, and the searcher selects from the candidate words. A candidate word is adopted and added to a search sentence, and a candidate word following the word adopted in the search sentence is presented and the candidate word selected by the searcher is added to the search sentence a required number of times. Is created, a search target document is searched with the inclusion of the search sentence as a search condition, and a process of presenting the search result is executed.

The recording medium 25 is a floppy (registered trademark) disk, a CD-ROM, a magneto-optical disk, a semiconductor memory, or the like in which a program for executing the search preprocessing method or the document search method is stored.

<Other Embodiments> In the above embodiment, an example in which a search is performed with one search sentence has been described. However, a plurality of search sentences may be created and configured to perform an AND search or an OR search. good.

In the above embodiment, an example is shown in which candidate words following a candidate word used in a search sentence are presented.
The present invention is not limited to this, and a preceding candidate word may be presented.

In addition, when presenting candidate words, words having different half-width and full-width characters, differences between katakana and hiragana, presence or absence of long sounds, and the like may be presented in the same row, and synonyms and related words may be displayed using a thesaurus. You may comprise so that the candidate word which is a word may also be shown.

The above-described retrieval pre-processing device and the document retrieval device may be provided separately or integrally. Further, the search pre-processing device and the document search device may target documents stored in a device such as the Internet via a communication line. For example, documents stored in each server on the Internet cover various fields, and new words and coined words are sometimes used. Therefore, trial and error are often required for searching. Since the candidate word is created from the above, an appropriate search sentence can be easily created even in such a case, which is particularly useful. In addition, since the present invention creates a search sentence while referring to candidate words, there are few omissions in the search, and a wide variety of documents such as documents collected by questionnaires, questions and answers to technical support, online help of computer software, etc. It is suitable when searching for appropriate information without omission.

As described above, according to the present invention, a natural search sentence is created using candidate words, and a search is performed.
Make it easy to create appropriate search statements,
A search preprocessing device, a document search device, a search preprocessing method, and a document search method that enable a search in which a search intention is appropriately reflected can be provided.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram of a first embodiment according to the present invention.

FIG. 2 is a flowchart of a search procedure according to the first embodiment.

FIG. 3 is a diagram showing a display example at the time of document search;

FIG. 4 is a diagram showing a display example at the time of document search.

FIG. 5 is a diagram showing a display example at the time of document search.

FIG. 6 is a diagram showing a display example at the time of document search.

FIG. 7 is a diagram showing a display example at the time of document search.

FIG. 8 shows a search target document.

FIG. 9 is an explanatory diagram of a separation process.

FIG. 10 is an explanatory diagram of a candidate word table.

FIG. 11 is an explanatory diagram of a frequency table.

FIG. 12 is an explanatory diagram of a keyword table.

FIG. 13 is a schematic configuration diagram of a second embodiment according to the present invention.

FIG. 14 is a flowchart of a search procedure according to the second embodiment.

FIG. 15 is a diagram showing a display example at the time of document search.

FIG. 16 is a diagram showing a display example at the time of document search.

FIG. 17 is a diagram showing a display example at the time of document search.

FIG. 18 is an explanatory diagram of an automatic classification procedure.

FIG. 19 is an explanatory diagram of a weighting process.

FIG. 20 is an explanatory diagram of weight data.

FIG. 21 is an explanatory diagram of a candidate word table.

FIG. 22 is an explanatory diagram in the case of considering appearance forms of a plurality of candidate words.

FIG. 23 is a schematic configuration diagram of a fourth embodiment according to the present invention.

FIG. 24 is an explanatory diagram of creating a combination of keywords.

FIG. 25 is an explanatory diagram of creating a combination of search sentences.

FIG. 26 is a schematic configuration diagram of a fifth embodiment according to the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Search preprocessing apparatus 1a Candidate word creation means 1b Accumulation means 1c, 1c 'Keyword extraction means 1d Weighting means 1e Combination creation means 2 Document search apparatus 2a Candidate word presentation means 2b Search sentence creation means 2c Search result presentation means 10 Database 20 Computer 21 Input unit 22 Storage unit 23 Arithmetic processing unit 24 Display unit 25 Recording medium

Continuing on the front page (72) Inventor Keizo Uchiyama Tokyo Electric Power Company, Inc. 1-3-1, Uchisaiwai-cho, Chiyoda-ku, Tokyo (72) Inventor Itsuko Izuko 1-3-1, Uchisaiwaicho, Chiyoda-ku, Tokyo Tokyo Electric Power Stock Inside the company (72) Inventor Akio Yasuda 6-14-1, Nishi-Shinjuku, Shinjuku-ku, Tokyo Shinjuku Green Tower Building 19F Peace Information Center Co., Ltd. (72) Inventor Misa Onuma 6-14-1 Nishi-Shinjuku, Shinjuku-ku, Tokyo Green Tower Building 19F Peace Information Center Co., Ltd. (72) Inventor Naoko Makita 6-14-1 Nishi-Shinjuku, Shinjuku-ku, Tokyo Shinjuku Green Tower Building 19F Peace Information Center Co., Ltd. F-term (reference) 5B075 ND03 NK02 NK31 PP24 PQ02 PQ46 PQ75 PR04

Claims (17)

[Claims] 1. A document to be searched is divided into words of a predetermined unit,
A search preprocessing apparatus comprising: candidate word creating means for creating a candidate word serving as a candidate for a search condition; and candidate word storage means for storing the candidate word in association with the appearance form of the candidate word. . 2. The method according to claim 1, further comprising: integrating means for integrating the number of occurrences of the same candidate word among the candidate words created by the candidate word creating means; and number-of-times storage means for storing the number of occurrences for each candidate word. The retrieval pre-processing device according to claim 1, wherein: 3. The apparatus according to claim 1, further comprising: candidate word weighting means for performing weighting according to the importance of the candidate word; and weight storage means for storing weighting information for each of the candidate words. 3. The search preprocessing device according to 2. 4. The pre-search processing apparatus according to claim 3, wherein the importance of the candidate word is at least one of the number of appearances and the location of the candidate word. 5. The apparatus according to claim 1, further comprising: keyword extracting means for extracting a keyword from the candidate words; and keyword storing means for storing the keyword.
The search preprocessing device according to 3 or 4. 6. A combination creating means for creating a combination of the keywords present in the search target document, and combination storage means for storing the combination of the keywords in association with the document in which the combination exists. The search preprocessing device according to claim 5, wherein 7. A document search apparatus for searching a plurality of documents as a search target and searching for a document that matches a search condition, comprising: candidate word presenting means for presenting a candidate word as a search condition candidate; A search sentence creating means for creating a search sentence by combining the candidate words presented to the means, and a search result presenting means for presenting a result of searching for the search target with the inclusion of the search sentence as a search condition. A document search device characterized by the following. 8. The search sentence creating unit according to claim 7, wherein, among the candidate words presented by the candidate word presenting unit, a candidate word selected by a searcher is adopted in the search sentence. Document retrieval device. 9. The method according to claim 1, wherein the candidate word presenting unit is configured to divide the search target document into words of a predetermined unit and generate the search word based on the appearance of the candidate word in the search target document. The candidate word following the adopted candidate word is presented to the creating unit, and the search sentence creating unit adopts the candidate word selected by the searcher among the following candidate words and adds the candidate word to the search sentence. 9. The document search device according to claim 8, wherein: 10. The candidate word presenting means, in accordance with the order of a plurality of candidate words adopted prior to the search sentence creating means, sets a next succeeding candidate word in the candidate word search target document. The document search apparatus according to claim 9, wherein the document is determined and presented based on an appearance form of the document. 11. The document search device according to claim 9, wherein the candidate word presenting means presents the candidate words in an order according to the number of appearances of the candidate words in the search document. 12. The document search apparatus according to claim 9, wherein the candidate word presenting means presents the candidate words in an order according to the importance of the candidate words. 13. The document search device according to claim 9, wherein a search sentence is formed by combining a noun or a noun phrase among a plurality of candidate words adopted by the search sentence creating means. . 14. A plurality of candidate words adopted by the search sentence creating means, a noun or a noun phrase is combined into a search sentence, and the search result presenting means divides the search target document into words in a predetermined unit. The combination of keywords that match the search sentence is searched for by referring to the combination of keywords that are nouns or noun phrases among the candidate words created in the above, and a document in which the combination of keywords that match this search sentence exists as a search result 14. The document search device according to claim 9, wherein the document is presented. 15. The method according to claim 1, wherein the search target document is divided into words of a predetermined unit, candidate words serving as search condition candidates are created, and the candidate words are stored in association with the appearance forms of the candidate words. Search preprocessing method. 16. A candidate word which is a candidate for a search condition is presented, and a candidate word selected by a searcher from among the candidate words is adopted as a first word in a search sentence, and a word adopted in the search sentence is provided. , The candidate word selected by the searcher from the candidate words is adopted and added to the search sentence, and the candidate word following the word adopted in the search sentence is presented to the searcher. A search sentence is created by repeating the addition of the selected candidate word to the search sentence a required number of times, a search target document is searched using the search sentence being included as a search condition, and a search result is presented. The document search method you want. 17. A candidate word is created by dividing a search target document into words in a predetermined unit, and the candidate word is stored together with an appearance form in the search target document, and the search text is stored based on the appearance form. 17. The document search method according to claim 16, wherein a candidate word following the word adopted in step (c) is determined.